Prompting Basics

In the last article we covered some of the key limitations of LLMs. You can find the article here. In this guide we dive deeper into what prompting how to get high quality response from your LLM.

You have almost certainly used ChatGPT, Claude, Gemini, or some other LLM by now and have seen variations in response quality. Perhaps you asked it to draft an email and have seen first hand that sometimes the answer is brilliant but at other times it is vague, wrong or misses the mark. You may be surprise to read that actually a lot of this comes down to the instruction you originally gave the LLM.

That instruction you type into the chat box is called a prompt. And the skill of writing better instructions is called prompting. This two-part guide covers what prompting is, why it matters, and (in Part 2) the practical techniques that will make you noticeably better at it.

If you want to understand more about what is happening under the bonnet of these tools, and where they tend to fall short, have a look at our earlier article on the limitations of large language models. That context is useful here, because many prompting techniques exist specifically to work around those limitations.

What is a prompt?

A prompt is simply the input you give an AI model to get a response. It could be:

• A question: "What is the best way to onboard a new employee remotely?"
• An instruction: "Write me a thank-you email to a client."
• A conversation: "I am planning a trip to Lisbon. Can you suggest a one-week itinerary for someone who likes food and architecture?"

Everything you type into the chat box is your prompt. The AI reads it, interprets it, and generates a response based on what you have written.

Think of it like giving a brief to a very capable (and very eager) intern who knows nothing about your situation, your preferences, or your business. They can do excellent work, but only if you tell them what you actually need.

Why does it matter

You may think that surely AI is at the stage where it can take my simple question and run with it, but the truth of the matter is very similar to the intern analogy, the better instrucitons you give it, the better output you'll get from it. It will also be able to tweak the responses to your specific requirements.

Here is a quick example:

Weak prompt: "Tell me about marketing."

Stronger prompt: "I run a small bakery in Manchester, England. Can you explain three low-cost marketing strategies I could use to attract more local customers? Keep the language simple and give a brief example for each."

The first prompt could return a 2,000-word essay on the history of marketing. The second is far more likely to give you something useful and relevant.

Research consistently shows that specific, well-structured prompts can improve the relevance and accuracy of AI responses by a significant margin. Clear instructions, context, and constraints all help. Vague instructions leave the AI guessing, and guessing leads to generic answers (and hallucinations).

Garbage In -> Garbage Out

This is an old computer science saying which applies equally to the data and AI worlds. "Garbage in, garbage out" means the quality of the response you get back from the AI is directly linked to the quality of the prompt you give it. A low quality (too short, vague or inaccurate) prompt may give you the right answer, but it's also likely to give you an answer that misses the mark in terms of tone, content or style.

High-quality prompting allows you to:

1. Save time: You don't have to rewrite the AI's work multiple times.
2. Unlock creativity: AI can brainstorm ideas you would never think of, if you push it in the right direction.
3. Handle complex tasks: You can ask AI to analyse spreadsheets, summarize 50-page reports, or write code, but only if you explain the task clearly.

What makes a good prompt

This is the single most important thing to understand. These tools work by predicting the most likely helpful response based on what you have written. They do not know:

• Who you are: Are you a student, a CEO or a retired teacher?
• What you already know: Are you a novice or an expert in the subject?
• What format you want: A bullet list? A formal letter? A one paragraph summary?
• What does "good" look like: One person's perfect answer is another person's first draft.

You need to load all of this information into the prompt and get the AI primed.

Prompting is essentially the act of providing the AI with a comprehensive set of instructions so it can have the best chance of giving you the output you want. The more clearly you do that, the better the result.

Three ideas worth understanding before Part 2

1. Context is everything

Context means the background information that helps the AI understand your situation. Without it, the AI has to guess.

Without context: "Write a cover letter."

With context: "Write a cover letter for a marketing assistant role at a charity. I have two years of experience in social media management and a degree in English. The tone should be professional but warm. Use British English."

You do not need to write an essay. Just give the AI the essential details it needs to tailor its response.

2. Iteration is normal

Very few people write the perfect prompt on their first try, and that is completely fine. Working with AI is usually best thought of as a conversation, not a single command.

You can ask a question, read the response, and then refine your request. You might say "that is too formal, can you make it more conversational?" or "good, but I forgot to mention it also needs to include our opening hours." This back-and-forth is called iteration, and it is how most people get the best results. You do not need to get everything right up front. However, giving the LLM good instructions means fewer cycles of iteration and better final outcome.

3. Be specific

The more specific you are about what you want, the closer the output will be to what you actually need. Vague instructions leave room for the AI to interpret things differently from what you intended. Specific instructions narrow the possibilities and point the AI in the right direction.

For example, something like the below will give you much better results:

"I am learning about the history of artificial intelligence. Write me a non-technical introduction on key milestones in the field. I'd also like references to key papers and research so I can look up source material. I want this to be comprehensive and a roadmap for my learning. If you use technical terms, provide me with definitions. This response should be written to Markdown file so I can read it offline."

A quick word on expectations

Rather than treat AI as a black box, think of it as a tool that you can use but requires mastering. To get the best results and to hone your tool you should:

• Review the output: AI can make mistakes, present outdated information, or confidently state something that is not true ("hallucinate").
• Use your own judgement: You know your audience, your business, and your situation better than any AI does.
• Treat it as a starting point: Think of AI output as a strong first draft which you should review and refine.

Good prompting dramatically improves what you get back, but it does not remove the need for a human eye.

Wrapping up

To summarise:

• A prompt is the instruction or question you give to an AI tool.
• The quality of your prompt directly affects the quality of the response.
• AI models do not know anything about you unless you provide that context.
• Iteration (refining your prompt through conversation) is a normal and effective way to work.
• Specificity gives you big improvements.

In Part 2, we will get into the practical techniques: how to structure your prompts, common patterns that work well, and mistakes to avoid. If you have ever felt like AI tools are not quite living up to the hype, better prompting is almost always the answer.